lustre/lov/lov_request.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2013, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  */
  36
  37 #define DEBUG_SUBSYSTEM S_LOV
  38
  39 #ifdef __KERNEL__
  40 #include <libcfs/libcfs.h>
  41 #else
  42 #include <liblustre.h>
  43 #endif
  44
  45 #include <obd_class.h>
  46 #include <obd_lov.h>
  47 #include <lustre/lustre_idl.h>
  48
  49 #include "lov_internal.h"
  50
  51 static void lov_init_set(struct lov_request_set *set)
  52 {
  53         set->set_count = 0;
  54         cfs_atomic_set(&set->set_completes, 0);
  55         cfs_atomic_set(&set->set_success, 0);
  56         cfs_atomic_set(&set->set_finish_checked, 0);
  57         set->set_cookies = 0;
  58         CFS_INIT_LIST_HEAD(&set->set_list);
  59         cfs_atomic_set(&set->set_refcount, 1);
  60         cfs_waitq_init(&set->set_waitq);
  61         spin_lock_init(&set->set_lock);
  62 }
  63
  64 void lov_finish_set(struct lov_request_set *set)
  65 {
  66         cfs_list_t *pos, *n;
  67         ENTRY;
  68
  69         LASSERT(set);
  70         cfs_list_for_each_safe(pos, n, &set->set_list) {
  71                 struct lov_request *req = cfs_list_entry(pos,
  72                                                          struct lov_request,
  73                                                          rq_link);
  74                 cfs_list_del_init(&req->rq_link);
  75
  76                 if (req->rq_oi.oi_oa)
  77                         OBDO_FREE(req->rq_oi.oi_oa);
  78                 if (req->rq_oi.oi_md)
  79                         OBD_FREE_LARGE(req->rq_oi.oi_md, req->rq_buflen);
  80                 if (req->rq_oi.oi_osfs)
  81                         OBD_FREE(req->rq_oi.oi_osfs,
  82                                  sizeof(*req->rq_oi.oi_osfs));
  83                 OBD_FREE(req, sizeof(*req));
  84         }
  85
  86         if (set->set_pga) {
  87                 int len = set->set_oabufs * sizeof(*set->set_pga);
  88                 OBD_FREE_LARGE(set->set_pga, len);
  89         }
  90         if (set->set_lockh)
  91                 lov_llh_put(set->set_lockh);
  92
  93         OBD_FREE(set, sizeof(*set));
  94         EXIT;
  95 }
  96
  97 int lov_set_finished(struct lov_request_set *set, int idempotent)
  98 {
  99         int completes = cfs_atomic_read(&set->set_completes);
 100
 101         CDEBUG(D_INFO, "check set %d/%d\n", completes, set->set_count);
 102
 103         if (completes == set->set_count) {
 104                 if (idempotent)
 105                         return 1;
 106                 if (cfs_atomic_inc_return(&set->set_finish_checked) == 1)
 107                         return 1;
 108         }
 109         return 0;
 110 }
 111
 112 void lov_update_set(struct lov_request_set *set,
 113                     struct lov_request *req, int rc)
 114 {
 115         req->rq_complete = 1;
 116         req->rq_rc = rc;
 117
 118         cfs_atomic_inc(&set->set_completes);
 119         if (rc == 0)
 120                 cfs_atomic_inc(&set->set_success);
 121
 122         cfs_waitq_signal(&set->set_waitq);
 123 }
 124
 125 int lov_update_common_set(struct lov_request_set *set,
 126                           struct lov_request *req, int rc)
 127 {
 128         struct lov_obd *lov = &set->set_exp->exp_obd->u.lov;
 129         ENTRY;
 130
 131         lov_update_set(set, req, rc);
 132
 133         /* grace error on inactive ost */
 134         if (rc && !(lov->lov_tgts[req->rq_idx] &&
 135                     lov->lov_tgts[req->rq_idx]->ltd_active))
 136                 rc = 0;
 137
 138         /* FIXME in raid1 regime, should return 0 */
 139         RETURN(rc);
 140 }
 141
 142 void lov_set_add_req(struct lov_request *req, struct lov_request_set *set)
 143 {
 144         cfs_list_add_tail(&req->rq_link, &set->set_list);
 145         set->set_count++;
 146         req->rq_rqset = set;
 147 }
 148
 149 static int lov_check_set(struct lov_obd *lov, int idx)
 150 {
 151         int rc = 0;
 152         mutex_lock(&lov->lov_lock);
 153
 154         if (lov->lov_tgts[idx] == NULL ||
 155             lov->lov_tgts[idx]->ltd_active ||
 156             (lov->lov_tgts[idx]->ltd_exp != NULL &&
 157              class_exp2cliimp(lov->lov_tgts[idx]->ltd_exp)->imp_connect_tried))
 158                 rc = 1;
 159
 160         mutex_unlock(&lov->lov_lock);
 161         return rc;
 162 }
 163
 164 /* Check if the OSC connection exists and is active.
 165  * If the OSC has not yet had a chance to connect to the OST the first time,
 166  * wait once for it to connect instead of returning an error.
 167  */
 168 int lov_check_and_wait_active(struct lov_obd *lov, int ost_idx)
 169 {
 170         cfs_waitq_t waitq;
 171         struct l_wait_info lwi;
 172         struct lov_tgt_desc *tgt;
 173         int rc = 0;
 174
 175         mutex_lock(&lov->lov_lock);
 176
 177         tgt = lov->lov_tgts[ost_idx];
 178
 179         if (unlikely(tgt == NULL))
 180                 GOTO(out, rc = 0);
 181
 182         if (likely(tgt->ltd_active))
 183                 GOTO(out, rc = 1);
 184
 185         if (tgt->ltd_exp && class_exp2cliimp(tgt->ltd_exp)->imp_connect_tried)
 186                 GOTO(out, rc = 0);
 187
 188         mutex_unlock(&lov->lov_lock);
 189
 190         cfs_waitq_init(&waitq);
 191         lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(obd_timeout),
 192                                    cfs_time_seconds(1), NULL, NULL);
 193
 194         rc = l_wait_event(waitq, lov_check_set(lov, ost_idx), &lwi);
 195         if (tgt != NULL && tgt->ltd_active)
 196                 return 1;
 197
 198         return 0;
 199
 200 out:
 201         mutex_unlock(&lov->lov_lock);
 202         return rc;
 203 }
 204
 205 extern void osc_update_enqueue(struct lustre_handle *lov_lockhp,
 206                                struct lov_oinfo *loi, int flags,
 207                                struct ost_lvb *lvb, __u32 mode, int rc);
 208
 209 static int lov_update_enqueue_lov(struct obd_export *exp,
 210                                   struct lustre_handle *lov_lockhp,
 211                                   struct lov_oinfo *loi, int flags, int idx,
 212                                   struct ost_id *oi, int rc)
 213 {
 214         struct lov_obd *lov = &exp->exp_obd->u.lov;
 215
 216         if (rc != ELDLM_OK &&
 217             !(rc == ELDLM_LOCK_ABORTED && (flags & LDLM_FL_HAS_INTENT))) {
 218                 memset(lov_lockhp, 0, sizeof(*lov_lockhp));
 219                 if (lov->lov_tgts[idx] && lov->lov_tgts[idx]->ltd_active) {
 220                         /* -EUSERS used by OST to report file contention */
 221                         if (rc != -EINTR && rc != -EUSERS)
 222                                 CERROR("%s: enqueue objid "DOSTID" subobj"
 223                                        DOSTID" on OST idx %d: rc %d\n",
 224                                        exp->exp_obd->obd_name,
 225                                        POSTID(oi), POSTID(&loi->loi_oi),
 226                                        loi->loi_ost_idx, rc);
 227                 } else
 228                         rc = ELDLM_OK;
 229         }
 230         return rc;
 231 }
 232
 233 int lov_update_enqueue_set(struct lov_request *req, __u32 mode, int rc)
 234 {
 235         struct lov_request_set *set = req->rq_rqset;
 236         struct lustre_handle *lov_lockhp;
 237         struct obd_info *oi = set->set_oi;
 238         struct lov_oinfo *loi;
 239         ENTRY;
 240
 241         LASSERT(oi != NULL);
 242
 243         lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
 244         loi = oi->oi_md->lsm_oinfo[req->rq_stripe];
 245
 246         /* XXX LOV STACKING: OSC gets a copy, created in lov_prep_enqueue_set
 247          * and that copy can be arbitrarily out of date.
 248          *
 249          * The LOV API is due for a serious rewriting anyways, and this
 250          * can be addressed then. */
 251
 252         lov_stripe_lock(oi->oi_md);
 253         osc_update_enqueue(lov_lockhp, loi, oi->oi_flags,
 254                            &req->rq_oi.oi_md->lsm_oinfo[0]->loi_lvb, mode, rc);
 255         if (rc == ELDLM_LOCK_ABORTED && (oi->oi_flags & LDLM_FL_HAS_INTENT))
 256                 memset(lov_lockhp, 0, sizeof *lov_lockhp);
 257         rc = lov_update_enqueue_lov(set->set_exp, lov_lockhp, loi, oi->oi_flags,
 258                                     req->rq_idx, &oi->oi_md->lsm_oi, rc);
 259         lov_stripe_unlock(oi->oi_md);
 260         lov_update_set(set, req, rc);
 261         RETURN(rc);
 262 }
 263
 264 /* The callback for osc_enqueue that updates lov info for every OSC request. */
 265 static int cb_update_enqueue(void *cookie, int rc)
 266 {
 267         struct obd_info *oinfo = cookie;
 268         struct ldlm_enqueue_info *einfo;
 269         struct lov_request *lovreq;
 270
 271         lovreq = container_of(oinfo, struct lov_request, rq_oi);
 272         einfo = lovreq->rq_rqset->set_ei;
 273         return lov_update_enqueue_set(lovreq, einfo->ei_mode, rc);
 274 }
 275
 276 static int enqueue_done(struct lov_request_set *set, __u32 mode)
 277 {
 278         struct lov_request *req;
 279         struct lov_obd *lov = &set->set_exp->exp_obd->u.lov;
 280         int completes = cfs_atomic_read(&set->set_completes);
 281         int rc = 0;
 282         ENTRY;
 283
 284         /* enqueue/match success, just return */
 285         if (completes && completes == cfs_atomic_read(&set->set_success))
 286                 RETURN(0);
 287
 288         /* cancel enqueued/matched locks */
 289         cfs_list_for_each_entry(req, &set->set_list, rq_link) {
 290                 struct lustre_handle *lov_lockhp;
 291
 292                 if (!req->rq_complete || req->rq_rc)
 293                         continue;
 294
 295                 lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
 296                 LASSERT(lov_lockhp);
 297                 if (!lustre_handle_is_used(lov_lockhp))
 298                         continue;
 299
 300                 rc = obd_cancel(lov->lov_tgts[req->rq_idx]->ltd_exp,
 301                                 req->rq_oi.oi_md, mode, lov_lockhp);
 302                 if (rc && lov->lov_tgts[req->rq_idx] &&
 303                     lov->lov_tgts[req->rq_idx]->ltd_active)
 304                         CERROR("%s: cancelling obdjid "DOSTID" on OST"
 305                                "idx %d error: rc = %d\n",
 306                                set->set_exp->exp_obd->obd_name,
 307                                POSTID(&req->rq_oi.oi_md->lsm_oi),
 308                                req->rq_idx, rc);
 309         }
 310         if (set->set_lockh)
 311                 lov_llh_put(set->set_lockh);
 312         RETURN(rc);
 313 }
 314
 315 int lov_fini_enqueue_set(struct lov_request_set *set, __u32 mode, int rc,
 316                          struct ptlrpc_request_set *rqset)
 317 {
 318         int ret = 0;
 319         ENTRY;
 320
 321         if (set == NULL)
 322                 RETURN(0);
 323         LASSERT(set->set_exp);
 324         /* Do enqueue_done only for sync requests and if any request
 325          * succeeded. */
 326         if (!rqset) {
 327                 if (rc)
 328                         cfs_atomic_set(&set->set_completes, 0);
 329                 ret = enqueue_done(set, mode);
 330         } else if (set->set_lockh)
 331                 lov_llh_put(set->set_lockh);
 332
 333         lov_put_reqset(set);
 334
 335         RETURN(rc ? rc : ret);
 336 }
 337
 338 static void lov_llh_addref(void *llhp)
 339 {
 340         struct lov_lock_handles *llh = llhp;
 341
 342         cfs_atomic_inc(&llh->llh_refcount);
 343         CDEBUG(D_INFO, "GETting llh %p : new refcount %d\n", llh,
 344                cfs_atomic_read(&llh->llh_refcount));
 345 }
 346
 347 static struct portals_handle_ops lov_handle_ops = {
 348         .hop_addref = lov_llh_addref,
 349         .hop_free   = NULL,
 350 };
 351
 352 static struct lov_lock_handles *lov_llh_new(struct lov_stripe_md *lsm)
 353 {
 354         struct lov_lock_handles *llh;
 355
 356         OBD_ALLOC(llh, sizeof *llh +
 357                   sizeof(*llh->llh_handles) * lsm->lsm_stripe_count);
 358         if (llh == NULL)
 359                 return NULL;
 360
 361         cfs_atomic_set(&llh->llh_refcount, 2);
 362         llh->llh_stripe_count = lsm->lsm_stripe_count;
 363         CFS_INIT_LIST_HEAD(&llh->llh_handle.h_link);
 364         class_handle_hash(&llh->llh_handle, &lov_handle_ops);
 365
 366         return llh;
 367 }
 368
 369 int lov_prep_enqueue_set(struct obd_export *exp, struct obd_info *oinfo,
 370                          struct ldlm_enqueue_info *einfo,
 371                          struct lov_request_set **reqset)
 372 {
 373         struct lov_obd *lov = &exp->exp_obd->u.lov;
 374         struct lov_request_set *set;
 375         int i, rc = 0;
 376         ENTRY;
 377
 378         OBD_ALLOC(set, sizeof(*set));
 379         if (set == NULL)
 380                 RETURN(-ENOMEM);
 381         lov_init_set(set);
 382
 383         set->set_exp = exp;
 384         set->set_oi = oinfo;
 385         set->set_ei = einfo;
 386         set->set_lockh = lov_llh_new(oinfo->oi_md);
 387         if (set->set_lockh == NULL)
 388                 GOTO(out_set, rc = -ENOMEM);
 389         oinfo->oi_lockh->cookie = set->set_lockh->llh_handle.h_cookie;
 390
 391         for (i = 0; i < oinfo->oi_md->lsm_stripe_count; i++) {
 392                 struct lov_oinfo *loi;
 393                 struct lov_request *req;
 394                 obd_off start, end;
 395
 396                 loi = oinfo->oi_md->lsm_oinfo[i];
 397                 if (!lov_stripe_intersects(oinfo->oi_md, i,
 398                                            oinfo->oi_policy.l_extent.start,
 399                                            oinfo->oi_policy.l_extent.end,
 400                                            &start, &end))
 401                         continue;
 402
 403                 if (!lov_check_and_wait_active(lov, loi->loi_ost_idx)) {
 404                         CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
 405                         continue;
 406                 }
 407
 408                 OBD_ALLOC(req, sizeof(*req));
 409                 if (req == NULL)
 410                         GOTO(out_set, rc = -ENOMEM);
 411
 412                 req->rq_buflen = sizeof(*req->rq_oi.oi_md) +
 413                         sizeof(struct lov_oinfo *) +
 414                         sizeof(struct lov_oinfo);
 415                 OBD_ALLOC_LARGE(req->rq_oi.oi_md, req->rq_buflen);
 416                 if (req->rq_oi.oi_md == NULL) {
 417                         OBD_FREE(req, sizeof(*req));
 418                         GOTO(out_set, rc = -ENOMEM);
 419                 }
 420                 req->rq_oi.oi_md->lsm_oinfo[0] =
 421                         ((void *)req->rq_oi.oi_md) + sizeof(*req->rq_oi.oi_md) +
 422                         sizeof(struct lov_oinfo *);
 423
 424                 /* Set lov request specific parameters. */
 425                 req->rq_oi.oi_lockh = set->set_lockh->llh_handles + i;
 426                 req->rq_oi.oi_cb_up = cb_update_enqueue;
 427                 req->rq_oi.oi_flags = oinfo->oi_flags;
 428
 429                 LASSERT(req->rq_oi.oi_lockh);
 430
 431                 req->rq_oi.oi_policy.l_extent.gid =
 432                         oinfo->oi_policy.l_extent.gid;
 433                 req->rq_oi.oi_policy.l_extent.start = start;
 434                 req->rq_oi.oi_policy.l_extent.end = end;
 435
 436                 req->rq_idx = loi->loi_ost_idx;
 437                 req->rq_stripe = i;
 438
 439                 /* XXX LOV STACKING: submd should be from the subobj */
 440                 req->rq_oi.oi_md->lsm_oi = loi->loi_oi;
 441                 req->rq_oi.oi_md->lsm_stripe_count = 0;
 442                 req->rq_oi.oi_md->lsm_oinfo[0]->loi_kms_valid =
 443                         loi->loi_kms_valid;
 444                 req->rq_oi.oi_md->lsm_oinfo[0]->loi_kms = loi->loi_kms;
 445                 req->rq_oi.oi_md->lsm_oinfo[0]->loi_lvb = loi->loi_lvb;
 446
 447                 lov_set_add_req(req, set);
 448         }
 449         if (!set->set_count)
 450                 GOTO(out_set, rc = -EIO);
 451         *reqset = set;
 452         RETURN(0);
 453 out_set:
 454         lov_fini_enqueue_set(set, einfo->ei_mode, rc, NULL);
 455         RETURN(rc);
 456 }
 457
 458 int lov_fini_match_set(struct lov_request_set *set, __u32 mode, int flags)
 459 {
 460         int rc = 0;
 461         ENTRY;
 462
 463         if (set == NULL)
 464                 RETURN(0);
 465         LASSERT(set->set_exp);
 466         rc = enqueue_done(set, mode);
 467         if ((set->set_count == cfs_atomic_read(&set->set_success)) &&
 468             (flags & LDLM_FL_TEST_LOCK))
 469                 lov_llh_put(set->set_lockh);
 470
 471         lov_put_reqset(set);
 472
 473         RETURN(rc);
 474 }
 475
 476 int lov_prep_match_set(struct obd_export *exp, struct obd_info *oinfo,
 477                        struct lov_stripe_md *lsm, ldlm_policy_data_t *policy,
 478                        __u32 mode, struct lustre_handle *lockh,
 479                        struct lov_request_set **reqset)
 480 {
 481         struct lov_obd *lov = &exp->exp_obd->u.lov;
 482         struct lov_request_set *set;
 483         int i, rc = 0;
 484         ENTRY;
 485
 486         OBD_ALLOC(set, sizeof(*set));
 487         if (set == NULL)
 488                 RETURN(-ENOMEM);
 489         lov_init_set(set);
 490
 491         set->set_exp = exp;
 492         set->set_oi = oinfo;
 493         set->set_oi->oi_md = lsm;
 494         set->set_lockh = lov_llh_new(lsm);
 495         if (set->set_lockh == NULL)
 496                 GOTO(out_set, rc = -ENOMEM);
 497         lockh->cookie = set->set_lockh->llh_handle.h_cookie;
 498
 499         for (i = 0; i < lsm->lsm_stripe_count; i++){
 500                 struct lov_oinfo *loi;
 501                 struct lov_request *req;
 502                 obd_off start, end;
 503
 504                 loi = lsm->lsm_oinfo[i];
 505                 if (!lov_stripe_intersects(lsm, i, policy->l_extent.start,
 506                                            policy->l_extent.end, &start, &end))
 507                         continue;
 508
 509                 /* FIXME raid1 should grace this error */
 510                 if (!lov_check_and_wait_active(lov, loi->loi_ost_idx)) {
 511                         CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
 512                         GOTO(out_set, rc = -EIO);
 513                 }
 514
 515                 OBD_ALLOC(req, sizeof(*req));
 516                 if (req == NULL)
 517                         GOTO(out_set, rc = -ENOMEM);
 518
 519                 req->rq_buflen = sizeof(*req->rq_oi.oi_md);
 520                 OBD_ALLOC_LARGE(req->rq_oi.oi_md, req->rq_buflen);
 521                 if (req->rq_oi.oi_md == NULL) {
 522                         OBD_FREE(req, sizeof(*req));
 523                         GOTO(out_set, rc = -ENOMEM);
 524                 }
 525
 526                 req->rq_oi.oi_policy.l_extent.start = start;
 527                 req->rq_oi.oi_policy.l_extent.end = end;
 528                 req->rq_oi.oi_policy.l_extent.gid = policy->l_extent.gid;
 529
 530                 req->rq_idx = loi->loi_ost_idx;
 531                 req->rq_stripe = i;
 532
 533                 /* XXX LOV STACKING: submd should be from the subobj */
 534                 req->rq_oi.oi_md->lsm_oi = loi->loi_oi;
 535                 req->rq_oi.oi_md->lsm_stripe_count = 0;
 536
 537                 lov_set_add_req(req, set);
 538         }
 539         if (!set->set_count)
 540                 GOTO(out_set, rc = -EIO);
 541         *reqset = set;
 542         RETURN(rc);
 543 out_set:
 544         lov_fini_match_set(set, mode, 0);
 545         RETURN(rc);
 546 }
 547
 548 int lov_fini_cancel_set(struct lov_request_set *set)
 549 {
 550         int rc = 0;
 551         ENTRY;
 552
 553         if (set == NULL)
 554                 RETURN(0);
 555
 556         LASSERT(set->set_exp);
 557         if (set->set_lockh)
 558                 lov_llh_put(set->set_lockh);
 559
 560         lov_put_reqset(set);
 561
 562         RETURN(rc);
 563 }
 564
 565 int lov_prep_cancel_set(struct obd_export *exp, struct obd_info *oinfo,
 566                         struct lov_stripe_md *lsm, __u32 mode,
 567                         struct lustre_handle *lockh,
 568                         struct lov_request_set **reqset)
 569 {
 570         struct lov_request_set *set;
 571         int i, rc = 0;
 572         ENTRY;
 573
 574         OBD_ALLOC(set, sizeof(*set));
 575         if (set == NULL)
 576                 RETURN(-ENOMEM);
 577         lov_init_set(set);
 578
 579         set->set_exp = exp;
 580         set->set_oi = oinfo;
 581         set->set_oi->oi_md = lsm;
 582         set->set_lockh = lov_handle2llh(lockh);
 583         if (set->set_lockh == NULL) {
 584                 CERROR("LOV: invalid lov lock handle %p\n", lockh);
 585                 GOTO(out_set, rc = -EINVAL);
 586         }
 587         lockh->cookie = set->set_lockh->llh_handle.h_cookie;
 588
 589         for (i = 0; i < lsm->lsm_stripe_count; i++){
 590                 struct lov_request *req;
 591                 struct lustre_handle *lov_lockhp;
 592                 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
 593
 594                 lov_lockhp = set->set_lockh->llh_handles + i;
 595                 if (!lustre_handle_is_used(lov_lockhp)) {
 596                         CDEBUG(D_INFO, "lov idx %d subobj "DOSTID" no lock\n",
 597                                loi->loi_ost_idx, POSTID(&loi->loi_oi));
 598                         continue;
 599                 }
 600
 601                 OBD_ALLOC(req, sizeof(*req));
 602                 if (req == NULL)
 603                         GOTO(out_set, rc = -ENOMEM);
 604
 605                 req->rq_buflen = sizeof(*req->rq_oi.oi_md);
 606                 OBD_ALLOC_LARGE(req->rq_oi.oi_md, req->rq_buflen);
 607                 if (req->rq_oi.oi_md == NULL) {
 608                         OBD_FREE(req, sizeof(*req));
 609                         GOTO(out_set, rc = -ENOMEM);
 610                 }
 611
 612                 req->rq_idx = loi->loi_ost_idx;
 613                 req->rq_stripe = i;
 614
 615                 /* XXX LOV STACKING: submd should be from the subobj */
 616                 req->rq_oi.oi_md->lsm_oi = loi->loi_oi;
 617                 req->rq_oi.oi_md->lsm_stripe_count = 0;
 618
 619                 lov_set_add_req(req, set);
 620         }
 621         if (!set->set_count)
 622                 GOTO(out_set, rc = -EIO);
 623         *reqset = set;
 624         RETURN(rc);
 625 out_set:
 626         lov_fini_cancel_set(set);
 627         RETURN(rc);
 628 }
 629 static int common_attr_done(struct lov_request_set *set)
 630 {
 631         cfs_list_t *pos;
 632         struct lov_request *req;
 633         struct obdo *tmp_oa;
 634         int rc = 0, attrset = 0;
 635         ENTRY;
 636
 637         LASSERT(set->set_oi != NULL);
 638
 639         if (set->set_oi->oi_oa == NULL)
 640                 RETURN(0);
 641
 642         if (!cfs_atomic_read(&set->set_success))
 643                 RETURN(-EIO);
 644
 645         OBDO_ALLOC(tmp_oa);
 646         if (tmp_oa == NULL)
 647                 GOTO(out, rc = -ENOMEM);
 648
 649         cfs_list_for_each (pos, &set->set_list) {
 650                 req = cfs_list_entry(pos, struct lov_request, rq_link);
 651
 652                 if (!req->rq_complete || req->rq_rc)
 653                         continue;
 654                 if (req->rq_oi.oi_oa->o_valid == 0)   /* inactive stripe */
 655                         continue;
 656                 lov_merge_attrs(tmp_oa, req->rq_oi.oi_oa,
 657                                 req->rq_oi.oi_oa->o_valid,
 658                                 set->set_oi->oi_md, req->rq_stripe, &attrset);
 659         }
 660         if (!attrset) {
 661                 CERROR("No stripes had valid attrs\n");
 662                 rc = -EIO;
 663         }
 664         if ((set->set_oi->oi_oa->o_valid & OBD_MD_FLEPOCH) &&
 665             (set->set_oi->oi_md->lsm_stripe_count != attrset)) {
 666                 /* When we take attributes of some epoch, we require all the
 667                  * ost to be active. */
 668                 CERROR("Not all the stripes had valid attrs\n");
 669                 GOTO(out, rc = -EIO);
 670         }
 671
 672         tmp_oa->o_oi = set->set_oi->oi_oa->o_oi;
 673         memcpy(set->set_oi->oi_oa, tmp_oa, sizeof(*set->set_oi->oi_oa));
 674 out:
 675         if (tmp_oa)
 676                 OBDO_FREE(tmp_oa);
 677         RETURN(rc);
 678
 679 }
 680
 681 static int brw_done(struct lov_request_set *set)
 682 {
 683         struct lov_stripe_md *lsm = set->set_oi->oi_md;
 684         struct lov_oinfo     *loi = NULL;
 685         cfs_list_t *pos;
 686         struct lov_request *req;
 687         ENTRY;
 688
 689         cfs_list_for_each (pos, &set->set_list) {
 690                 req = cfs_list_entry(pos, struct lov_request, rq_link);
 691
 692                 if (!req->rq_complete || req->rq_rc)
 693                         continue;
 694
 695                 loi = lsm->lsm_oinfo[req->rq_stripe];
 696
 697                 if (req->rq_oi.oi_oa->o_valid & OBD_MD_FLBLOCKS)
 698                         loi->loi_lvb.lvb_blocks = req->rq_oi.oi_oa->o_blocks;
 699         }
 700
 701         RETURN(0);
 702 }
 703
 704 int lov_fini_brw_set(struct lov_request_set *set)
 705 {
 706         int rc = 0;
 707         ENTRY;
 708
 709         if (set == NULL)
 710                 RETURN(0);
 711         LASSERT(set->set_exp);
 712         if (cfs_atomic_read(&set->set_completes)) {
 713                 rc = brw_done(set);
 714                 /* FIXME update qos data here */
 715         }
 716         lov_put_reqset(set);
 717
 718         RETURN(rc);
 719 }
 720
 721 int lov_prep_brw_set(struct obd_export *exp, struct obd_info *oinfo,
 722                      obd_count oa_bufs, struct brw_page *pga,
 723                      struct obd_trans_info *oti,
 724                      struct lov_request_set **reqset)
 725 {
 726         struct {
 727                 obd_count       index;
 728                 obd_count       count;
 729                 obd_count       off;
 730         } *info = NULL;
 731         struct lov_request_set *set;
 732         struct lov_obd *lov = &exp->exp_obd->u.lov;
 733         int rc = 0, i, shift;
 734         ENTRY;
 735
 736         OBD_ALLOC(set, sizeof(*set));
 737         if (set == NULL)
 738                 RETURN(-ENOMEM);
 739         lov_init_set(set);
 740
 741         set->set_exp = exp;
 742         set->set_oti = oti;
 743         set->set_oi = oinfo;
 744         set->set_oabufs = oa_bufs;
 745         OBD_ALLOC_LARGE(set->set_pga, oa_bufs * sizeof(*set->set_pga));
 746         if (!set->set_pga)
 747                 GOTO(out, rc = -ENOMEM);
 748
 749         OBD_ALLOC_LARGE(info, sizeof(*info) * oinfo->oi_md->lsm_stripe_count);
 750         if (!info)
 751                 GOTO(out, rc = -ENOMEM);
 752
 753         /* calculate the page count for each stripe */
 754         for (i = 0; i < oa_bufs; i++) {
 755                 int stripe = lov_stripe_number(oinfo->oi_md, pga[i].off);
 756                 info[stripe].count++;
 757         }
 758
 759         /* alloc and initialize lov request */
 760         shift = 0;
 761         for (i = 0; i < oinfo->oi_md->lsm_stripe_count; i++){
 762                 struct lov_oinfo *loi = NULL;
 763                 struct lov_request *req;
 764
 765                 if (info[i].count == 0)
 766                         continue;
 767
 768                 loi = oinfo->oi_md->lsm_oinfo[i];
 769                 if (!lov_check_and_wait_active(lov, loi->loi_ost_idx)) {
 770                         CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
 771                         GOTO(out, rc = -EIO);
 772                 }
 773
 774                 OBD_ALLOC(req, sizeof(*req));
 775                 if (req == NULL)
 776                         GOTO(out, rc = -ENOMEM);
 777
 778                 OBDO_ALLOC(req->rq_oi.oi_oa);
 779                 if (req->rq_oi.oi_oa == NULL) {
 780                         OBD_FREE(req, sizeof(*req));
 781                         GOTO(out, rc = -ENOMEM);
 782                 }
 783
 784                 if (oinfo->oi_oa) {
 785                         memcpy(req->rq_oi.oi_oa, oinfo->oi_oa,
 786                                sizeof(*req->rq_oi.oi_oa));
 787                 }
 788                 req->rq_oi.oi_oa->o_oi = loi->loi_oi;
 789                 req->rq_oi.oi_oa->o_stripe_idx = i;
 790
 791                 req->rq_buflen = sizeof(*req->rq_oi.oi_md);
 792                 OBD_ALLOC_LARGE(req->rq_oi.oi_md, req->rq_buflen);
 793                 if (req->rq_oi.oi_md == NULL) {
 794                         OBDO_FREE(req->rq_oi.oi_oa);
 795                         OBD_FREE(req, sizeof(*req));
 796                         GOTO(out, rc = -ENOMEM);
 797                 }
 798
 799                 req->rq_idx = loi->loi_ost_idx;
 800                 req->rq_stripe = i;
 801
 802                 /* XXX LOV STACKING */
 803                 req->rq_oi.oi_md->lsm_oi = loi->loi_oi;
 804                 req->rq_oabufs = info[i].count;
 805                 req->rq_pgaidx = shift;
 806                 shift += req->rq_oabufs;
 807
 808                 /* remember the index for sort brw_page array */
 809                 info[i].index = req->rq_pgaidx;
 810
 811                 req->rq_oi.oi_capa = oinfo->oi_capa;
 812
 813                 lov_set_add_req(req, set);
 814         }
 815         if (!set->set_count)
 816                 GOTO(out, rc = -EIO);
 817
 818         /* rotate & sort the brw_page array */
 819         for (i = 0; i < oa_bufs; i++) {
 820                 int stripe = lov_stripe_number(oinfo->oi_md, pga[i].off);
 821
 822                 shift = info[stripe].index + info[stripe].off;
 823                 LASSERT(shift < oa_bufs);
 824                 set->set_pga[shift] = pga[i];
 825                 lov_stripe_offset(oinfo->oi_md, pga[i].off, stripe,
 826                                   &set->set_pga[shift].off);
 827                 info[stripe].off++;
 828         }
 829 out:
 830         if (info)
 831                 OBD_FREE_LARGE(info,
 832                                sizeof(*info) * oinfo->oi_md->lsm_stripe_count);
 833
 834         if (rc == 0)
 835                 *reqset = set;
 836         else
 837                 lov_fini_brw_set(set);
 838
 839         RETURN(rc);
 840 }
 841
 842 int lov_fini_getattr_set(struct lov_request_set *set)
 843 {
 844         int rc = 0;
 845         ENTRY;
 846
 847         if (set == NULL)
 848                 RETURN(0);
 849         LASSERT(set->set_exp);
 850         if (cfs_atomic_read(&set->set_completes))
 851                 rc = common_attr_done(set);
 852
 853         lov_put_reqset(set);
 854
 855         RETURN(rc);
 856 }
 857
 858 /* The callback for osc_getattr_async that finilizes a request info when a
 859  * response is received. */
 860 static int cb_getattr_update(void *cookie, int rc)
 861 {
 862         struct obd_info *oinfo = cookie;
 863         struct lov_request *lovreq;
 864         lovreq = container_of(oinfo, struct lov_request, rq_oi);
 865         return lov_update_common_set(lovreq->rq_rqset, lovreq, rc);
 866 }
 867
 868 int lov_prep_getattr_set(struct obd_export *exp, struct obd_info *oinfo,
 869                          struct lov_request_set **reqset)
 870 {
 871         struct lov_request_set *set;
 872         struct lov_obd *lov = &exp->exp_obd->u.lov;
 873         int rc = 0, i;
 874         ENTRY;
 875
 876         OBD_ALLOC(set, sizeof(*set));
 877         if (set == NULL)
 878                 RETURN(-ENOMEM);
 879         lov_init_set(set);
 880
 881         set->set_exp = exp;
 882         set->set_oi = oinfo;
 883
 884         for (i = 0; i < oinfo->oi_md->lsm_stripe_count; i++) {
 885                 struct lov_oinfo *loi;
 886                 struct lov_request *req;
 887
 888                 loi = oinfo->oi_md->lsm_oinfo[i];
 889                 if (!lov_check_and_wait_active(lov, loi->loi_ost_idx)) {
 890                         CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
 891                         if (oinfo->oi_oa->o_valid & OBD_MD_FLEPOCH)
 892                                 /* SOM requires all the OSTs to be active. */
 893                                 GOTO(out_set, rc = -EIO);
 894                         continue;
 895                 }
 896
 897                 OBD_ALLOC(req, sizeof(*req));
 898                 if (req == NULL)
 899                         GOTO(out_set, rc = -ENOMEM);
 900
 901                 req->rq_stripe = i;
 902                 req->rq_idx = loi->loi_ost_idx;
 903
 904                 OBDO_ALLOC(req->rq_oi.oi_oa);
 905                 if (req->rq_oi.oi_oa == NULL) {
 906                         OBD_FREE(req, sizeof(*req));
 907                         GOTO(out_set, rc = -ENOMEM);
 908                 }
 909                 memcpy(req->rq_oi.oi_oa, oinfo->oi_oa,
 910                        sizeof(*req->rq_oi.oi_oa));
 911                 req->rq_oi.oi_oa->o_oi = loi->loi_oi;
 912                 req->rq_oi.oi_cb_up = cb_getattr_update;
 913                 req->rq_oi.oi_capa = oinfo->oi_capa;
 914
 915                 lov_set_add_req(req, set);
 916         }
 917         if (!set->set_count)
 918                 GOTO(out_set, rc = -EIO);
 919         *reqset = set;
 920         RETURN(rc);
 921 out_set:
 922         lov_fini_getattr_set(set);
 923         RETURN(rc);
 924 }
 925
 926 int lov_fini_destroy_set(struct lov_request_set *set)
 927 {
 928         ENTRY;
 929
 930         if (set == NULL)
 931                 RETURN(0);
 932         LASSERT(set->set_exp);
 933         if (cfs_atomic_read(&set->set_completes)) {
 934                 /* FIXME update qos data here */
 935         }
 936
 937         lov_put_reqset(set);
 938
 939         RETURN(0);
 940 }
 941
 942 int lov_prep_destroy_set(struct obd_export *exp, struct obd_info *oinfo,
 943                          struct obdo *src_oa, struct lov_stripe_md *lsm,
 944                          struct obd_trans_info *oti,
 945                          struct lov_request_set **reqset)
 946 {
 947         struct lov_request_set *set;
 948         struct lov_obd *lov = &exp->exp_obd->u.lov;
 949         int rc = 0, i;
 950         ENTRY;
 951
 952         OBD_ALLOC(set, sizeof(*set));
 953         if (set == NULL)
 954                 RETURN(-ENOMEM);
 955         lov_init_set(set);
 956
 957         set->set_exp = exp;
 958         set->set_oi = oinfo;
 959         set->set_oi->oi_md = lsm;
 960         set->set_oi->oi_oa = src_oa;
 961         set->set_oti = oti;
 962         if (oti != NULL && src_oa->o_valid & OBD_MD_FLCOOKIE)
 963                 set->set_cookies = oti->oti_logcookies;
 964
 965         for (i = 0; i < lsm->lsm_stripe_count; i++) {
 966                 struct lov_oinfo *loi;
 967                 struct lov_request *req;
 968
 969                 loi = lsm->lsm_oinfo[i];
 970                 if (!lov_check_and_wait_active(lov, loi->loi_ost_idx)) {
 971                         CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
 972                         continue;
 973                 }
 974
 975                 OBD_ALLOC(req, sizeof(*req));
 976                 if (req == NULL)
 977                         GOTO(out_set, rc = -ENOMEM);
 978
 979                 req->rq_stripe = i;
 980                 req->rq_idx = loi->loi_ost_idx;
 981
 982                 OBDO_ALLOC(req->rq_oi.oi_oa);
 983                 if (req->rq_oi.oi_oa == NULL) {
 984                         OBD_FREE(req, sizeof(*req));
 985                         GOTO(out_set, rc = -ENOMEM);
 986                 }
 987                 memcpy(req->rq_oi.oi_oa, src_oa, sizeof(*req->rq_oi.oi_oa));
 988                 req->rq_oi.oi_oa->o_oi = loi->loi_oi;
 989                 lov_set_add_req(req, set);
 990         }
 991         if (!set->set_count)
 992                 GOTO(out_set, rc = -EIO);
 993         *reqset = set;
 994         RETURN(rc);
 995 out_set:
 996         lov_fini_destroy_set(set);
 997         RETURN(rc);
 998 }
 999
1000 int lov_fini_setattr_set(struct lov_request_set *set)
1001 {
1002         int rc = 0;
1003         ENTRY;
1004
1005         if (set == NULL)
1006                 RETURN(0);
1007         LASSERT(set->set_exp);
1008         if (cfs_atomic_read(&set->set_completes)) {
1009                 rc = common_attr_done(set);
1010                 /* FIXME update qos data here */
1011         }
1012
1013         lov_put_reqset(set);
1014         RETURN(rc);
1015 }
1016
1017 int lov_update_setattr_set(struct lov_request_set *set,
1018                            struct lov_request *req, int rc)
1019 {
1020         struct lov_obd *lov = &req->rq_rqset->set_exp->exp_obd->u.lov;
1021         struct lov_stripe_md *lsm = req->rq_rqset->set_oi->oi_md;
1022         ENTRY;
1023
1024         lov_update_set(set, req, rc);
1025
1026         /* grace error on inactive ost */
1027         if (rc && !(lov->lov_tgts[req->rq_idx] &&
1028                     lov->lov_tgts[req->rq_idx]->ltd_active))
1029                 rc = 0;
1030
1031         if (rc == 0) {
1032                 if (req->rq_oi.oi_oa->o_valid & OBD_MD_FLCTIME)
1033                         lsm->lsm_oinfo[req->rq_stripe]->loi_lvb.lvb_ctime =
1034                                 req->rq_oi.oi_oa->o_ctime;
1035                 if (req->rq_oi.oi_oa->o_valid & OBD_MD_FLMTIME)
1036                         lsm->lsm_oinfo[req->rq_stripe]->loi_lvb.lvb_mtime =
1037                                 req->rq_oi.oi_oa->o_mtime;
1038                 if (req->rq_oi.oi_oa->o_valid & OBD_MD_FLATIME)
1039                         lsm->lsm_oinfo[req->rq_stripe]->loi_lvb.lvb_atime =
1040                                 req->rq_oi.oi_oa->o_atime;
1041         }
1042
1043         RETURN(rc);
1044 }
1045
1046 /* The callback for osc_setattr_async that finilizes a request info when a
1047  * response is received. */
1048 static int cb_setattr_update(void *cookie, int rc)
1049 {
1050         struct obd_info *oinfo = cookie;
1051         struct lov_request *lovreq;
1052         lovreq = container_of(oinfo, struct lov_request, rq_oi);
1053         return lov_update_setattr_set(lovreq->rq_rqset, lovreq, rc);
1054 }
1055
1056 int lov_prep_setattr_set(struct obd_export *exp, struct obd_info *oinfo,
1057                          struct obd_trans_info *oti,
1058                          struct lov_request_set **reqset)
1059 {
1060         struct lov_request_set *set;
1061         struct lov_obd *lov = &exp->exp_obd->u.lov;
1062         int rc = 0, i;
1063         ENTRY;
1064
1065         OBD_ALLOC(set, sizeof(*set));
1066         if (set == NULL)
1067                 RETURN(-ENOMEM);
1068         lov_init_set(set);
1069
1070         set->set_exp = exp;
1071         set->set_oti = oti;
1072         set->set_oi = oinfo;
1073         if (oti != NULL && oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
1074                 set->set_cookies = oti->oti_logcookies;
1075
1076         for (i = 0; i < oinfo->oi_md->lsm_stripe_count; i++) {
1077                 struct lov_oinfo *loi = oinfo->oi_md->lsm_oinfo[i];
1078                 struct lov_request *req;
1079
1080                 if (!lov_check_and_wait_active(lov, loi->loi_ost_idx)) {
1081                         CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
1082                         continue;
1083                 }
1084
1085                 OBD_ALLOC(req, sizeof(*req));
1086                 if (req == NULL)
1087                         GOTO(out_set, rc = -ENOMEM);
1088                 req->rq_stripe = i;
1089                 req->rq_idx = loi->loi_ost_idx;
1090
1091                 OBDO_ALLOC(req->rq_oi.oi_oa);
1092                 if (req->rq_oi.oi_oa == NULL) {
1093                         OBD_FREE(req, sizeof(*req));
1094                         GOTO(out_set, rc = -ENOMEM);
1095                 }
1096                 memcpy(req->rq_oi.oi_oa, oinfo->oi_oa,
1097                        sizeof(*req->rq_oi.oi_oa));
1098                 req->rq_oi.oi_oa->o_oi = loi->loi_oi;
1099                 req->rq_oi.oi_oa->o_stripe_idx = i;
1100                 req->rq_oi.oi_cb_up = cb_setattr_update;
1101                 req->rq_oi.oi_capa = oinfo->oi_capa;
1102
1103                 if (oinfo->oi_oa->o_valid & OBD_MD_FLSIZE) {
1104                         int off = lov_stripe_offset(oinfo->oi_md,
1105                                                     oinfo->oi_oa->o_size, i,
1106                                                     &req->rq_oi.oi_oa->o_size);
1107
1108                         if (off < 0 && req->rq_oi.oi_oa->o_size)
1109                                 req->rq_oi.oi_oa->o_size--;
1110
1111                         CDEBUG(D_INODE, "stripe %d has size "LPU64"/"LPU64"\n",
1112                                i, req->rq_oi.oi_oa->o_size,
1113                                oinfo->oi_oa->o_size);
1114                 }
1115                 lov_set_add_req(req, set);
1116         }
1117         if (!set->set_count)
1118                 GOTO(out_set, rc = -EIO);
1119         *reqset = set;
1120         RETURN(rc);
1121 out_set:
1122         lov_fini_setattr_set(set);
1123         RETURN(rc);
1124 }
1125
1126 int lov_fini_punch_set(struct lov_request_set *set)
1127 {
1128         int rc = 0;
1129         ENTRY;
1130
1131         if (set == NULL)
1132                 RETURN(0);
1133         LASSERT(set->set_exp);
1134         if (cfs_atomic_read(&set->set_completes)) {
1135                 rc = -EIO;
1136                 /* FIXME update qos data here */
1137                 if (cfs_atomic_read(&set->set_success))
1138                         rc = common_attr_done(set);
1139         }
1140
1141         lov_put_reqset(set);
1142
1143         RETURN(rc);
1144 }
1145
1146 int lov_update_punch_set(struct lov_request_set *set,
1147                          struct lov_request *req, int rc)
1148 {
1149         struct lov_obd *lov = &req->rq_rqset->set_exp->exp_obd->u.lov;
1150         struct lov_stripe_md *lsm = req->rq_rqset->set_oi->oi_md;
1151         ENTRY;
1152
1153         lov_update_set(set, req, rc);
1154
1155         /* grace error on inactive ost */
1156         if (rc && !lov->lov_tgts[req->rq_idx]->ltd_active)
1157                 rc = 0;
1158
1159         if (rc == 0) {
1160                 lov_stripe_lock(lsm);
1161                 if (req->rq_oi.oi_oa->o_valid & OBD_MD_FLBLOCKS) {
1162                         lsm->lsm_oinfo[req->rq_stripe]->loi_lvb.lvb_blocks =
1163                                 req->rq_oi.oi_oa->o_blocks;
1164                 }
1165
1166                 lov_stripe_unlock(lsm);
1167         }
1168
1169         RETURN(rc);
1170 }
1171
1172 /* The callback for osc_punch that finilizes a request info when a response
1173  * is received. */
1174 static int cb_update_punch(void *cookie, int rc)
1175 {
1176         struct obd_info *oinfo = cookie;
1177         struct lov_request *lovreq;
1178         lovreq = container_of(oinfo, struct lov_request, rq_oi);
1179         return lov_update_punch_set(lovreq->rq_rqset, lovreq, rc);
1180 }
1181
1182 int lov_prep_punch_set(struct obd_export *exp, struct obd_info *oinfo,
1183                        struct obd_trans_info *oti,
1184                        struct lov_request_set **reqset)
1185 {
1186         struct lov_request_set *set;
1187         struct lov_obd *lov = &exp->exp_obd->u.lov;
1188         int rc = 0, i;
1189         ENTRY;
1190
1191         OBD_ALLOC(set, sizeof(*set));
1192         if (set == NULL)
1193                 RETURN(-ENOMEM);
1194         lov_init_set(set);
1195
1196         set->set_oi = oinfo;
1197         set->set_exp = exp;
1198
1199         for (i = 0; i < oinfo->oi_md->lsm_stripe_count; i++) {
1200                 struct lov_oinfo *loi = oinfo->oi_md->lsm_oinfo[i];
1201                 struct lov_request *req;
1202                 obd_off rs, re;
1203
1204                 if (!lov_stripe_intersects(oinfo->oi_md, i,
1205                                            oinfo->oi_policy.l_extent.start,
1206                                            oinfo->oi_policy.l_extent.end,
1207                                            &rs, &re))
1208                         continue;
1209
1210                 if (!lov_check_and_wait_active(lov, loi->loi_ost_idx)) {
1211                         CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
1212                         GOTO(out_set, rc = -EIO);
1213                 }
1214
1215                 OBD_ALLOC(req, sizeof(*req));
1216                 if (req == NULL)
1217                         GOTO(out_set, rc = -ENOMEM);
1218                 req->rq_stripe = i;
1219                 req->rq_idx = loi->loi_ost_idx;
1220
1221                 OBDO_ALLOC(req->rq_oi.oi_oa);
1222                 if (req->rq_oi.oi_oa == NULL) {
1223                         OBD_FREE(req, sizeof(*req));
1224                         GOTO(out_set, rc = -ENOMEM);
1225                 }
1226                 memcpy(req->rq_oi.oi_oa, oinfo->oi_oa,
1227                        sizeof(*req->rq_oi.oi_oa));
1228                 req->rq_oi.oi_oa->o_oi = loi->loi_oi;
1229                 req->rq_oi.oi_oa->o_valid |= OBD_MD_FLGROUP;
1230
1231                 req->rq_oi.oi_oa->o_stripe_idx = i;
1232                 req->rq_oi.oi_cb_up = cb_update_punch;
1233
1234                 req->rq_oi.oi_policy.l_extent.start = rs;
1235                 req->rq_oi.oi_policy.l_extent.end = re;
1236                 req->rq_oi.oi_policy.l_extent.gid = -1;
1237
1238                 req->rq_oi.oi_capa = oinfo->oi_capa;
1239
1240                 lov_set_add_req(req, set);
1241         }
1242         if (!set->set_count)
1243                 GOTO(out_set, rc = -EIO);
1244         *reqset = set;
1245         RETURN(rc);
1246 out_set:
1247         lov_fini_punch_set(set);
1248         RETURN(rc);
1249 }
1250
1251 int lov_fini_sync_set(struct lov_request_set *set)
1252 {
1253         int rc = 0;
1254         ENTRY;
1255
1256         if (set == NULL)
1257                 RETURN(0);
1258         LASSERT(set->set_exp);
1259         if (cfs_atomic_read(&set->set_completes)) {
1260                 if (!cfs_atomic_read(&set->set_success))
1261                         rc = -EIO;
1262                 /* FIXME update qos data here */
1263         }
1264
1265         lov_put_reqset(set);
1266
1267         RETURN(rc);
1268 }
1269
1270 /* The callback for osc_sync that finilizes a request info when a
1271  * response is recieved. */
1272 static int cb_sync_update(void *cookie, int rc)
1273 {
1274         struct obd_info *oinfo = cookie;
1275         struct lov_request *lovreq;
1276
1277         lovreq = container_of(oinfo, struct lov_request, rq_oi);
1278         return lov_update_common_set(lovreq->rq_rqset, lovreq, rc);
1279 }
1280
1281 int lov_prep_sync_set(struct obd_export *exp, struct obd_info *oinfo,
1282                       obd_off start, obd_off end,
1283                       struct lov_request_set **reqset)
1284 {
1285         struct lov_request_set *set;
1286         struct lov_obd *lov = &exp->exp_obd->u.lov;
1287         int rc = 0, i;
1288         ENTRY;
1289
1290         OBD_ALLOC_PTR(set);
1291         if (set == NULL)
1292                 RETURN(-ENOMEM);
1293         lov_init_set(set);
1294
1295         set->set_exp = exp;
1296         set->set_oi = oinfo;
1297
1298         for (i = 0; i < oinfo->oi_md->lsm_stripe_count; i++) {
1299                 struct lov_oinfo *loi = oinfo->oi_md->lsm_oinfo[i];
1300                 struct lov_request *req;
1301                 obd_off rs, re;
1302
1303                 if (!lov_check_and_wait_active(lov, loi->loi_ost_idx)) {
1304                         CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
1305                         continue;
1306                 }
1307
1308                 if (!lov_stripe_intersects(oinfo->oi_md, i, start, end, &rs,
1309                                            &re))
1310                         continue;
1311
1312                 OBD_ALLOC_PTR(req);
1313                 if (req == NULL)
1314                         GOTO(out_set, rc = -ENOMEM);
1315                 req->rq_stripe = i;
1316                 req->rq_idx = loi->loi_ost_idx;
1317
1318                 OBDO_ALLOC(req->rq_oi.oi_oa);
1319                 if (req->rq_oi.oi_oa == NULL) {
1320                         OBD_FREE(req, sizeof(*req));
1321                         GOTO(out_set, rc = -ENOMEM);
1322                 }
1323                 *req->rq_oi.oi_oa = *oinfo->oi_oa;
1324                 req->rq_oi.oi_oa->o_oi = loi->loi_oi;
1325                 req->rq_oi.oi_oa->o_stripe_idx = i;
1326
1327                 req->rq_oi.oi_policy.l_extent.start = rs;
1328                 req->rq_oi.oi_policy.l_extent.end = re;
1329                 req->rq_oi.oi_policy.l_extent.gid = -1;
1330                 req->rq_oi.oi_cb_up = cb_sync_update;
1331
1332                 lov_set_add_req(req, set);
1333         }
1334         if (!set->set_count)
1335                 GOTO(out_set, rc = -EIO);
1336         *reqset = set;
1337         RETURN(rc);
1338 out_set:
1339         lov_fini_sync_set(set);
1340         RETURN(rc);
1341 }
1342
1343 #define LOV_U64_MAX ((__u64)~0ULL)
1344 #define LOV_SUM_MAX(tot, add)                                           \
1345         do {                                                            \
1346                 if ((tot) + (add) < (tot))                              \
1347                         (tot) = LOV_U64_MAX;                            \
1348                 else                                                    \
1349                         (tot) += (add);                                 \
1350         } while(0)
1351
1352 int lov_fini_statfs(struct obd_device *obd, struct obd_statfs *osfs,int success)
1353 {
1354         ENTRY;
1355
1356         if (success) {
1357                 __u32 expected_stripes = lov_get_stripecnt(&obd->u.lov,
1358                                                            LOV_MAGIC, 0);
1359                 if (osfs->os_files != LOV_U64_MAX)
1360                         lov_do_div64(osfs->os_files, expected_stripes);
1361                 if (osfs->os_ffree != LOV_U64_MAX)
1362                         lov_do_div64(osfs->os_ffree, expected_stripes);
1363
1364                 spin_lock(&obd->obd_osfs_lock);
1365                 memcpy(&obd->obd_osfs, osfs, sizeof(*osfs));
1366                 obd->obd_osfs_age = cfs_time_current_64();
1367                 spin_unlock(&obd->obd_osfs_lock);
1368                 RETURN(0);
1369         }
1370
1371         RETURN(-EIO);
1372 }
1373
1374 int lov_fini_statfs_set(struct lov_request_set *set)
1375 {
1376         int rc = 0;
1377         ENTRY;
1378
1379         if (set == NULL)
1380                 RETURN(0);
1381
1382         if (cfs_atomic_read(&set->set_completes)) {
1383                 rc = lov_fini_statfs(set->set_obd, set->set_oi->oi_osfs,
1384                                      cfs_atomic_read(&set->set_success));
1385         }
1386         lov_put_reqset(set);
1387         RETURN(rc);
1388 }
1389
1390 void lov_update_statfs(struct obd_statfs *osfs, struct obd_statfs *lov_sfs,
1391                        int success)
1392 {
1393         int shift = 0, quit = 0;
1394         __u64 tmp;
1395
1396         if (success == 0) {
1397                 memcpy(osfs, lov_sfs, sizeof(*lov_sfs));
1398         } else {
1399                 if (osfs->os_bsize != lov_sfs->os_bsize) {
1400                         /* assume all block sizes are always powers of 2 */
1401                         /* get the bits difference */
1402                         tmp = osfs->os_bsize | lov_sfs->os_bsize;
1403                         for (shift = 0; shift <= 64; ++shift) {
1404                                 if (tmp & 1) {
1405                                         if (quit)
1406                                                 break;
1407                                         else
1408                                                 quit = 1;
1409                                         shift = 0;
1410                                 }
1411                                 tmp >>= 1;
1412                         }
1413                 }
1414
1415                 if (osfs->os_bsize < lov_sfs->os_bsize) {
1416                         osfs->os_bsize = lov_sfs->os_bsize;
1417
1418                         osfs->os_bfree  >>= shift;
1419                         osfs->os_bavail >>= shift;
1420                         osfs->os_blocks >>= shift;
1421                 } else if (shift != 0) {
1422                         lov_sfs->os_bfree  >>= shift;
1423                         lov_sfs->os_bavail >>= shift;
1424                         lov_sfs->os_blocks >>= shift;
1425                 }
1426 #ifdef MIN_DF
1427                 /* Sandia requested that df (and so, statfs) only
1428                    returned minimal available space on
1429                    a single OST, so people would be able to
1430                    write this much data guaranteed. */
1431                 if (osfs->os_bavail > lov_sfs->os_bavail) {
1432                         /* Presumably if new bavail is smaller,
1433                            new bfree is bigger as well */
1434                         osfs->os_bfree = lov_sfs->os_bfree;
1435                         osfs->os_bavail = lov_sfs->os_bavail;
1436                 }
1437 #else
1438                 osfs->os_bfree += lov_sfs->os_bfree;
1439                 osfs->os_bavail += lov_sfs->os_bavail;
1440 #endif
1441                 osfs->os_blocks += lov_sfs->os_blocks;
1442                 /* XXX not sure about this one - depends on policy.
1443                  *   - could be minimum if we always stripe on all OBDs
1444                  *     (but that would be wrong for any other policy,
1445                  *     if one of the OBDs has no more objects left)
1446                  *   - could be sum if we stripe whole objects
1447                  *   - could be average, just to give a nice number
1448                  *
1449                  * To give a "reasonable" (if not wholly accurate)
1450                  * number, we divide the total number of free objects
1451                  * by expected stripe count (watch out for overflow).
1452                  */
1453                 LOV_SUM_MAX(osfs->os_files, lov_sfs->os_files);
1454                 LOV_SUM_MAX(osfs->os_ffree, lov_sfs->os_ffree);
1455         }
1456 }
1457
1458 /* The callback for osc_statfs_async that finilizes a request info when a
1459  * response is received. */
1460 static int cb_statfs_update(void *cookie, int rc)
1461 {
1462         struct obd_info *oinfo = cookie;
1463         struct lov_request *lovreq;
1464         struct lov_request_set *set;
1465         struct obd_statfs *osfs, *lov_sfs;
1466         struct lov_obd *lov;
1467         struct lov_tgt_desc *tgt;
1468         struct obd_device *lovobd, *tgtobd;
1469         int success;
1470         ENTRY;
1471
1472         lovreq = container_of(oinfo, struct lov_request, rq_oi);
1473         set = lovreq->rq_rqset;
1474         lovobd = set->set_obd;
1475         lov = &lovobd->u.lov;
1476         osfs = set->set_oi->oi_osfs;
1477         lov_sfs = oinfo->oi_osfs;
1478         success = cfs_atomic_read(&set->set_success);
1479         /* XXX: the same is done in lov_update_common_set, however
1480            lovset->set_exp is not initialized. */
1481         lov_update_set(set, lovreq, rc);
1482         if (rc)
1483                 GOTO(out, rc);
1484
1485         obd_getref(lovobd);
1486         tgt = lov->lov_tgts[lovreq->rq_idx];
1487         if (!tgt || !tgt->ltd_active)
1488                 GOTO(out_update, rc);
1489
1490         tgtobd = class_exp2obd(tgt->ltd_exp);
1491         spin_lock(&tgtobd->obd_osfs_lock);
1492         memcpy(&tgtobd->obd_osfs, lov_sfs, sizeof(*lov_sfs));
1493         if ((oinfo->oi_flags & OBD_STATFS_FROM_CACHE) == 0)
1494                 tgtobd->obd_osfs_age = cfs_time_current_64();
1495         spin_unlock(&tgtobd->obd_osfs_lock);
1496
1497 out_update:
1498         lov_update_statfs(osfs, lov_sfs, success);
1499         obd_putref(lovobd);
1500
1501 out:
1502         if (set->set_oi->oi_flags & OBD_STATFS_PTLRPCD &&
1503             lov_set_finished(set, 0)) {
1504                 lov_statfs_interpret(NULL, set, set->set_count !=
1505                                      cfs_atomic_read(&set->set_success));
1506         }
1507
1508         RETURN(0);
1509 }
1510
1511 int lov_prep_statfs_set(struct obd_device *obd, struct obd_info *oinfo,
1512                         struct lov_request_set **reqset)
1513 {
1514         struct lov_request_set *set;
1515         struct lov_obd *lov = &obd->u.lov;
1516         int rc = 0, i;
1517         ENTRY;
1518
1519         OBD_ALLOC(set, sizeof(*set));
1520         if (set == NULL)
1521                 RETURN(-ENOMEM);
1522         lov_init_set(set);
1523
1524         set->set_obd = obd;
1525         set->set_oi = oinfo;
1526
1527         /* We only get block data from the OBD */
1528         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1529                 struct lov_request *req;
1530
1531                 if (lov->lov_tgts[i] == NULL ||
1532                     (!lov_check_and_wait_active(lov, i) &&
1533                      (oinfo->oi_flags & OBD_STATFS_NODELAY))) {
1534                         CDEBUG(D_HA, "lov idx %d inactive\n", i);
1535                         continue;
1536                 }
1537
1538                 /* skip targets that have been explicitely disabled by the
1539                  * administrator */
1540                 if (!lov->lov_tgts[i]->ltd_exp) {
1541                         CDEBUG(D_HA, "lov idx %d administratively disabled\n", i);
1542                         continue;
1543                 }
1544
1545                 OBD_ALLOC(req, sizeof(*req));
1546                 if (req == NULL)
1547                         GOTO(out_set, rc = -ENOMEM);
1548
1549                 OBD_ALLOC(req->rq_oi.oi_osfs, sizeof(*req->rq_oi.oi_osfs));
1550                 if (req->rq_oi.oi_osfs == NULL) {
1551                         OBD_FREE(req, sizeof(*req));
1552                         GOTO(out_set, rc = -ENOMEM);
1553                 }
1554
1555                 req->rq_idx = i;
1556                 req->rq_oi.oi_cb_up = cb_statfs_update;
1557                 req->rq_oi.oi_flags = oinfo->oi_flags;
1558
1559                 lov_set_add_req(req, set);
1560         }
1561         if (!set->set_count)
1562                 GOTO(out_set, rc = -EIO);
1563         *reqset = set;
1564         RETURN(rc);
1565 out_set:
1566         lov_fini_statfs_set(set);
1567         RETURN(rc);
1568 }